基于PCA-GWO-SVM的大坝变形预测

Dam Deformation Prediction Based on PCA-GWO-SVM Model

很多大坝失事前会出现坝体变形测值异常的情况,为了确保大坝安全运行,需要建立精确的模型来进行数据分析和变形预测。针对大坝安全监测数据序列中出现小样本、短序列和序列非线性的特点,将主成分分析法(PCA)引入支持向量机(SVM)来简化因子关系,实现支持向量机模型输入的优化设计,同时应用灰狼优化算法(GWO)对支持向量机进行参数优化,并结合支持向量机的非线性拟合能力,使模型更好地体现大坝的工作机制。以某混凝土连拱坝为例,分别建立统计模型、标准SVM模型、PCA-SVM模型以及PCA-GWO-SVM模型并对预测结果进行分析,对比验证了PCA-GWO-SVM模型方法的可行性。

It is very common for abnormal deformation measurements to occur before dam crash.Therefore,in order to ensure the safety oper⁃ation of the dam,accurate models are needed for data analysis and dam deformation prediction.In this paper,a combination model combining principal component analysis(PCA)and grey wolf optimization(GWO)algorithm with support vector machine(SVM)was established for the characteristics of small samples,short sequences and sequence non-linearity in the dam monitoring data sequence.It used PCA extracted the main features of the factors to simplify the factor relationship and to optimize the input of SVM model.At the same time,GWO algorithm was used to optimize the penalty coefficient and kernel function parameters of SVM model.Combined with the nonlinear fitting ability of sup⁃port vector machine,the model could better reflect the working mechanism of the dam.Taking a concrete multi-arch dam as an example,this paper established statistical model,standard SVM model,PCA-SVM model and PCA-GWO-SVM model respectively,and analyzed the pre⁃diction results to verify the feasibility of PCA-GWO-SVM model method.